The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: In SKILL.md, the instructions advise the user to install the uv tool using a shell script fetched via curl from https://astral.sh/uv/install.sh. While this is a standard installation method for a well-known developer tool, it involves piping remote content to a shell.
[COMMAND_EXECUTION]: The agent is directed to use the Bash tool to run a local script, scripts/extract_pdf.py, via the uv run command. This is used to extract trait values from assessment PDFs for further analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes assessment data from PDF files using OCR and image processing in scripts/culture_index/opencv_extractor.py. This creates a vulnerability surface where malicious content embedded in a PDF could be ingested into the agent context.
Ingestion points: PDF files are processed by scripts/extract_pdf.py.
Boundary markers: The extraction script converts PDF data into a structured JSON format, which acts as a data boundary, though the agent later interprets the content of these fields.
Capability inventory: The agent has access to Bash, Read, Grep, Glob, and Write tools, which could be abused if an injection is successful.
Sanitization: The script performs coordinate-based extraction for charts and OCR for metadata but does not explicitly sanitize the extracted text for prompt injection instructions.
[EXTERNAL_DOWNLOADS]: The extraction script declares dependencies on public Python packages (e.g., opencv-python-headless, numpy, pdf2image) which are automatically downloaded from public registries when the script is executed via the uv tool.

interpreting-culture-index